Description

Hypothesis

  • Mean of physiological and clinical features change differently over time in 4 primary groups - txp_pos, non_pos, txp_neg, non_neg. Our goal is to quantify the difference statistically.

Interpretation

  • If linear regression assumptions holds, the departure of confidence interval of feature mean conditioned/given a certain time in 4 groups reviews “significant” difference in time trend of feature mean in 4 groups.

Data

  • time window
    • (-24h,24h) around a blood culture
    • 48h window around an observation in a patient’s safe zone which has most number of data available
  • Imputation
    • raw

Medthod

  • Ordinary linear regression
    • Formula
      • feature ~ rcs(time, 5) * group
    • Cluster
      • multiple episodes from one patient
    • Outcome(y-axis)
      • the estimated conditional feature mean value with 95% confidence interval for the estimation
    • Assumptions
      • Linearity: responce feature is a linear combination of predictors
      • Independence: the errors are independent (conditioned on time)
      • Normality: the distribution of the errors should follow a normal distribution
      • Equal Variance: the error variance is the same at any set of predictors
    • Model diagnostics (model assumptions evaluation)
      • residual QQ-plot and histogram (check normal distribution assumption)
      • plot reesiduals against